
GeoPIFu: Geometry and Pixel Aligned Implicit Functions for Singleview Human Reconstruction
We propose GeoPIFu, a method to recover a 3D mesh from a monocular colo...
read it

Targeted Adversarial Perturbations for Monocular Depth Prediction
We study the effect of adversarial perturbations on the task of monocula...
read it

Stochastic batch size for adaptive regularization in deep network optimization
We propose a firstorder stochastic optimization algorithm incorporating...
read it

FDA: Fourier Domain Adaptation for Semantic Segmentation
We describe a simple method for unsupervised domain adaptation, whereby ...
read it

Learning to Manipulate Individual Objects in an Image
We describe a method to train a generative model with latent factors tha...
read it

Phase Consistent Ecological Domain Adaptation
We introduce two criteria to regularize the optimization involved in lea...
read it

Towards BackwardCompatible Representation Learning
We propose a way to learn visual features that are compatible with previ...
read it

Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from InputOutput Observations
We describe a procedure for removing dependency on a cohort of training ...
read it

Continual Universal Object Detection
Object detection has improved significantly in recent years on multiple ...
read it

Incremental Learning for MetricBased MetaLearners
Majority of the modern metalearning methods for fewshot classification...
read it

LaTeS: Latent Space Distillation for TeacherStudent Driving Policy Learning
We describe a policy learning approach to map visual inputs to driving c...
read it

Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Neural Networks
We explore the problem of selectively forgetting a particular set of dat...
read it

Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks
We explore the problem of selectively forgetting a particular set of dat...
read it

MetaQLearning
This paper introduces MetaQLearning (MQL), a new offpolicy algorithm ...
read it

A Baseline for FewShot Image Classification
Finetuning a deep network trained with the standard crossentropy loss ...
read it

Toward Understanding Catastrophic Forgetting in Continual Learning
We study the relationship between catastrophic forgetting and properties...
read it

Time Matters in Regularizing Deep Networks: Weight Decay and Data Augmentation Affect Early Learning Dynamics, Matter Little Near Convergence
Regularization is typically understood as improving generalization by al...
read it

Where is the Information in a Deep Neural Network?
Whatever information a Deep Neural Network has gleaned from past data is...
read it

Unsupervised Domain Adaptation via Regularized Conditional Alignment
We propose a method for unsupervised domain adaptation that trains a sha...
read it

VOICED: Depth Completion from Inertial Odometry and Vision
We describe a method to infer dense depth from camera motion and sparse ...
read it

FewShot Learning with Embedded Class Models and ShotFree Meta Training
We propose a method for learning embeddings for fewshot learning that i...
read it

MetaLearning with Differentiable Convex Optimization
Many metalearning approaches for fewshot learning rely on simple base ...
read it

The Information Complexity of Learning Tasks, their Structure and their Distance
We introduce an asymmetric distance in the space of learning tasks, and ...
read it

Bilateral Cyclic Constraint and Adaptive Regularization for Unsupervised Monocular Depth Prediction
Supervised learning methods to infer (hypothesize) depth of a scene from...
read it

Zero Shot Learning with the Isoperimetric Loss
We introduce the isoperimetric loss as a regularization criterion for le...
read it

Task2Vec: Task Embedding for MetaLearning
We introduce a method to provide vectorial representations of visual cla...
read it

Dense Depth Posterior (DDP) from Single Image and Sparse Range
We present a deep learning system to infer the posterior distribution of...
read it

Mono3D++: Monocular 3D Vehicle Detection with TwoScale 3D Hypotheses and Task Priors
We present a method to infer 3D pose and shape of vehicles from a single...
read it

Unsupervised Moving Object Detection via Contextual Information Separation
We propose an adversarial contextual model for detecting moving objects ...
read it

GeoNet: Deep Geodesic Networks for Point Cloud Analysis
Surfacebased geodesic topology provides strong cues for object semantic...
read it

The Dynamics of Differential Learning I: InformationDynamics and Task Reachability
We study the topology of the space of learning tasks, which is critical ...
read it

GeoSupervised Visual Depth Prediction
We propose using global orientation from inertial measurements, and the ...
read it

Conditional Prior Networks for Optical Flow
Classical computation of optical flow involves generic priors (regulariz...
read it

VisualInertial Object Detection and Mapping
We present a method to populate an unknown environment with models of pr...
read it

Input and Weight Space Smoothing for Semisupervised Learning
We propose regularizing the empirical loss for semisupervised learning ...
read it

SaaS: Speed as a Supervisor for Semisupervised Learning
We introduce the SaaS Algorithm for semisupervised learning, which uses...
read it

OATM: Occlusion Aware Template Matching by Consensus Set Maximization
We present a novel approach to template matching that is efficient, can ...
read it

Mathematics of Deep Learning
Recently there has been a dramatic increase in the performance of recogn...
read it

DeepRadiologyNet: Radiologist Level Pathology Detection in CT Head Images
We describe a system to automatically filter clinically significant find...
read it

Critical Learning Periods in Deep Neural Networks
Critical periods are phases in the early development of humans and anima...
read it

BlockCyclic Stochastic Coordinate Descent for Deep Neural Networks
We present a stochastic firstorder optimization algorithm, named BCSC, ...
read it

A Separation Principle for Control in the Age of Deep Learning
We review the problem of defining and inferring a "state" for a control ...
read it

Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks
Stochastic gradient descent (SGD) is widely believed to perform implicit...
read it

Parle: parallelizing stochastic gradient descent
We propose a new algorithm called Parle for parallel training of deep ne...
read it

Emergence of Invariance and Disentangling in Deep Representations
Using established principles from Information Theory and Statistics, we ...
read it

Person Depth ReID: Robust Person Reidentification with Commodity Depth Sensors
This work targets person reidentification (ReID) from depth sensors suc...
read it

Analysis of universal adversarial perturbations
Deep networks have recently been shown to be vulnerable to universal per...
read it

Classification regions of deep neural networks
The goal of this paper is to analyze the geometric properties of deep ne...
read it

Adaptive Regularization of Some Inverse Problems in Image Analysis
We present an adaptive regularization scheme for optimizing composite en...
read it

MultiLabel Segmentation via ResidualDriven Adaptive Regularization
We present a variational multilabel segmentation algorithm based on a r...
read it
Stefano Soatto
is this you? claim profile
Stefano Soatto is a Professor of Computer Science in Los Angeles, University of California, California, where he is also professor of electrical engineering, UCLA Vision Lab Founding Director. For his contributions to dynamic visual processes, he was named a Fellow of the Institute of Electrical and Electronic Engineers in 2013. In 1999 he was awarded the David Marr Prize in Computer Vision.
Soatto got his D. Eng. Eng. In 1992, in electric engineering Cum Laude was an EAP Fellow at Berkeley University in 199091, and received in 1996 his PhD in control and dynamic systems from the California Institute of Technology with his dissertation “A Geometrical Approach to Dynamic Vision.” In 1992, he completed his PhD at the University of California at Berkeley. In 199697, he was a postdoctoral scholar at Harvard University, and subsequently held positions at the university of Udine (Italy) as an Assistant and Associate Professor of Electrical Engineering and Biomedical Engineering at Washington University (St. Louis). Since 2000, he has been at UCLA.
Computer vision, machine teaching and robotics are the focus of Soatto’s research. He codeveloped optimal algorithms for From Motion Structure, characterized its ambiguities and also characterized the visual and inert fusion identity and observability. Its focus on research is the development of representations that are data functions that capture their informative content and eliminate irrelevant data variability.
Soatto’s lab demonstrated first in the CVPR 2000, ICCV 2001 and ECCV 2002 in realtime SFM and Augmented Reality on commodity hardware. He also coleaded the UCLAGolem team with Emilio Frazzoli and Amnon Shashua at DARPA’s second major autonomous vehicle challenge.